OpenAI unveils o1, a model that can fact-check itself

OpenAI's new o1 model can self-fact-check and excel in reasoning tasks.

: OpenAI has introduced o1, a new family of generative AI models, including o1-preview and o1-mini. These models are available on ChatGPT Plus or Team, with broader access planned. O1 features advanced reasoning abilities but is currently expensive and has limitations. It outperforms previous models in many tasks but sometimes hallucinates or takes longer to respond.

OpenAI has announced the launch of its new generative AI model family, o1, which includes o1-preview and o1-mini. These models are accessible via ChatGPT Plus or Team subscriptions, with enterprise and educational user access expected soon.

The o1 models have unique abilities to fact-check themselves and reason through complex tasks, which makes them stand out from previous versions like GPT-4o. Nevertheless, they come at a high cost and are not without flaws, such as occasional hallucinations and slower response times.

Despite these drawbacks, early tests indicate that o1 exceeds prior models in multiple areas, including coding and data analysis. OpenAI plans to further enhance o1's capabilities by experimenting with longer reasoning times, aiming to improve its performance even more.

Generative AIReinforcement LearningOpenAI